v1.0.0

Added

• ailoud audio import adds audio and video files, or whole directories, to a
  local library. --tag groups them so they can be found by context later.
• ailoud audio transcribe turns recordings into timestamped transcripts with
  whisper.cpp. --lang ru,en detects and preserves each language in a
  code-switched recording; --diarize attributes lines to speakers, and
  ailoud audio annotate --speaker speaker_00=Ann gives them real names.
• ailoud audio search finds where something was said across the library,
  returning the matching lines with timestamps rather than whole transcripts.
  Case-insensitive in every language, with * for a prefix search.
• ailoud audio summarize writes a summary with a language model and saves it
  as a report. --template shapes it for the kind of conversation -- a 1:1, a
  performance review, an architecture discussion, a decision between solutions
  -- and --context supplies what the transcript does not say. Templates are
  editable YAML files under the config directory. Four engines sit behind one
  setting: a local GGUF model through llama.cpp, Claude by subscription through
  the Claude Code CLI, Claude by API, and any OpenAI-compatible endpoint,
  including Ollama and LM Studio.
• ailoud report ls|show|rm lists, prints and deletes saved reports.
• ailoud mcp serves the library to an AI agent over MCP: sixteen tools, three
  prompts, and transcripts as addressable resources. Deleting takes two calls,
  the first describing exactly what would go.
• ailoud mcp install|uninstall|update configures Claude Code, Codex CLI,
  opencode, Gemini CLI, Hermes and GitHub Copilot CLI, and writes the guidance
  block those agents read.
• A project can keep its own library in .ailoud/, found by walking up from
  the working directory the way git finds .git.
• ailoud setup provisions a fresh machine -- ffmpeg, whisper.cpp, the models,
  and optionally a local summarisation model -- printing every command it will
  run, including any sudo, and asking once. ailoud doctor reports every
  missing tool, model and permission with a fix for each, and --fix acts on
  them.
• Commands are grouped by the noun they act on (ailoud audio ls,
  ailoud report show), each verb has a one-letter alias, and the short
  top-level spellings keep working.
• Output is decorated on a terminal and plain everywhere else, so --json and
  --format srt are byte-stable in a pipe. Anything over 30 lines opens in
  your pager.
• API keys are read from the environment only, never from the config file and
  never logged.
• Documentation at https://lorem-dev.github.io/ailoud/, published per release.
• Published to npm as ailoud, with @ailoud/core and @ailoud/providers for
  anyone building on the layers underneath.